Preliminary Analysis of Data

Figure2.6 illustrates the price trend in the market. Left panel depicts the evolution of monthly average price (in the form of a price index) by weighting each transaction with the volume. Right panel presents non-weighted values.

28_{Vickers and Ziebarth(2014) suggest that this change in price variation cannot be traced back} to factors of the cost as the cost variation is same before and after NIRA.

The price is relatively stable until month 7; then, it collapses without any sign of recovery. In a period of one year, it suffers a decline of almost 30 %. The average price in month 7–17 is 9.7% lower than the price average in the preceding months. This period is also marked with considerably higher variation in price. Standard deviation of price after month 7 is 34.9 % higher than the preceding period, while the coefficient of variation is 49.3% higher.

Figure 2.6: Monthly Price

Figure2.7illustrates the evolution of monthly average distance between provider and buyer. Month 7 marks the lowest average distance; after month 7, providers serve regions further away. Within a couple of months, the average distance increases 15 %. The direction of this expansion is also important. This is captured by the index of relative proximity measure, ∆j_l, which bases on the difference between provider distance to buyer fromthe distance of the closest rival to buyer. Similarly, it seems that initial stability in this measure is disrupted similarly in month 7. The index value is negative and increases in absolute value. Meaning that, firms extend their activities to regions closer to their rivals.

Figure 2.7: Distance of Operations

Serving customers closer to rivals necessitates price reductions. First panel in Figure 2.8 takes on this relationship. It illustrates the evolution of the correlation between transaction price and the distance of the closest rival to buyer each month. After month 7, the correlation between price and rival distance gradually increases;

expansion of operations is increasingly associated with price cuts. Second panel centres on the correlation between price and the number of rivals around the buyer in a defined radius; they are negatively correlated. Until month 7 the association of two variables shows a stable pattern. Similarly, the correlation increases in absolute terms in later months.

Figure 2.8: Price-Rival Distance & Price-Number of Rivals Correlation

Findings indicate that month 7 might represent a structural break in the form of regime switch from collusion to competition. In the next chapter, using proactive detection methodologies, this possibility is explored in greater detail.

Chapter 3

Proactive Detection: An Empirical

Application in a Spatial Setting

with Market Power Heterogeneity

3.1

Introduction

Competition policy draws its legal justification from competition law. Legal rules bans the undertakings from certain conducts, and defines the penalties they face if they do not comply. Within the borders of these prohibitions and penalties, the decision makers i.e. court, competition authority, evaluate the facts and the arguments of parties and form a decision. In a way, this decision is what gives cartel its existence; it is difficult to talk about a cartel without any legal ruling. Any legal ruling that say a cartel does/does not exit, builds on a prosecution; which requires willingness to investigate on the law enforcement side, which in turn depends on the available evidence. Some rare cases aside, the conventional wisdom in the competition policy is that economics have no role in the process of evidence gathering, or triggering an investigation. This is enshrined by the phrase “you can’t catch a thief with an economist1”. Rather, in “thief-catching”, policy makers around the world prefer relying on “the thieves themselves”. Today in many jurisdictions, leniency2_{is the primary tool to detect collusion. The role of the economist is confined} 1_{This phase is credited (Schinkel,2013, p.4) to Scott D. Hammond, Deputy Assistant Attorney} General for Criminal Enforcement in US Department of Justice. See, the presentation done by

Scott D. Hammond in October 2005, in OECD Prosecutor’s Programme Working Partyhttps://

www.justice.gov/atr/speech/ten-strategies-winning-fight-against-hardcore-cartels. 2_{Leniency is granted by the competition authority in exchange for cooperation. EU antitrust} regulations allow any firm that is part of a cartel to step up and acknowledge its participation in a collusive scheme; provide a detailed description of the collusive agreement i.e. coverage, duration, participants; and present any evidence it has. In return, if EU Commission does not have enough evidence prior to the application and the applicant is the first, an immunity from fine is granted. If the applicant firm is not the first to come forward, then instead of an immunity, a reduction is granted. The amount of reduction is 30–50 % for the second firm, 20–30% for the third firm, and

to the post ruling period; the assessment of the impact of an already proven collusion, which most probably is “detected” by a leniency application.

Not surprisingly, many economists have issues with this demarcation3_{. They}

find the overreliance to self-reporting troublesome; in particular, they are worried that leniency might not be desisting and deterring the cartels with highest member loyalty; these with highest profitability. Consequently, it is better to

complement leniency with other tools; to this end, they propose proactive detection methodologies, detecting collusive activity by using economic analysis in the absence of no prior information about the cartel.

Harrington (2008) describes the proposed methodology: The economist, similar to what a “detective” would do in pursuing a case, adopts a sequential analysis. The first step is“screening”. The aim in this stage is flagging markets that are “worthy of close scrutiny (p.214-215)”.

Building on collusive markers proposed byHarrington(2006a,c,2008), in Chapter 2, I conduct a simple analysis of data. Findings indicate that consistent with a regime switch from collusion to competition, stable relations in the market were disrupted after month seven. I take this as a suspicious pattern to be investigated further. In the sequential analysis ofHarrington (2008), if no suspicious pattern is identified in the screening, there is no basis for further investigation. If some suspicious patterns are identified, the law enforcement has two options. First, is triggering an ex-officio investigation with the available information4. Second, is asking the economist to go to the second stage, “the verification”. In this stage the aim is to

“systematically exclude competition as an explanation for observed behaviour and gather evidence in support of collusion”. The difference between screening and verification is this: Whereas screening may entail studying price patterns, verification requires controlling for demand and cost factors and any other variables necessary to distinguish between collusion and competition. (p.215)”. If the verification stage suggests patterns are consistent with competition, no further investigation is carried on. If not, authorities may trigger an ex-officio investigation. Harrington (2008) also suggests there might be a third stage, “the prosecution”; providing economic evidence“to persuade a court or some other administrative body that there has been violation of law”. This stage is essentially “verification with different standards (p.215)”, and is uncharted waters, as, in no jurisdiction, economic evidence is seen sufficient to establish guilt5_.

In this chapter, I take on the suspicious patterns identified around month seven,

20% for the subsequent firms. See, http://eur-lex.europa.eu/legal-content/EN/ALL/?uri=

CELEX:52006XC1208(04).

3_{See,Abrantes-Metz(2013b);Schinkel(2013).}

4_{At this point I presume that it is possible to trigger anex-officioinvestigation, which typically} would be more plausible if collusion is only an administrative offence. If collusion is a criminal offence, obtaining legal consent using economic evidence might be practically impossible.

5_{Note that this is different than damages proceedings where the impact of the conduct on price} is estimated, after the guilt is established by a court decision.

and investigate further, while I control for demand and cost shifters. Building on the theoretical framework laid on in Chapter 1, I explore if observed patterns are more consistent with collusion or competition.

In devising the empirical strategy, I consult to the empirical literature concerned with the identification of collusion. The literature in this area can be grouped into three: studies basing ondata analysis, works makingregime comparison, and studies

tracking collusive strategy. As covered in Chapter 2, data analysis is primarily used to identify suspicious markets, firms or periods. In some works, similar to this one, it is used as a first stage in a multi stage analysis. Regime comparison entails either (i) making estimations alternative regime assumptions and competing them in terms of likelihood, or (ii) estimating a single pricing/bidding equation assuming competition, and tracking the left over patterns in the unexplained portion. Tracking collusive strategy involves testing a specific type of correlation across the behaviour of rivals that arises as a result of the collusive strategy cartel employs or is suspected to employ.

One particularly important work in the literature for our purposes is Bresnahan (1987) who suggests that if there is price competition, for the products that have a close substitute, the price would converge to marginal cost; while in collusion, price and cost would diverge. He tests this theory in explaining drastic changes in US automobile industry in 1955, by taking this year as a temporary price war in the context of a longer collusion.

This study contributes first to the literatures of detecting collusion using consumer level data, and empirical analysis of price discrimination. However, more important contribution of this work is taking the premise in Bresnahan (1987) – centring on the relationship between price and local market power in identifying regime switch – that is applied to an heterogeneous product / product characteristics space setting to an homogeneous product / geographic space setting. In geographical space, local market power varies at each location according to cost difference between potential competitor and dominant competitor at that location. In this study, estimation centres on explaining pricing behaviour, and particularly its relation with ∆j_l, relative proximity of the provider and its closest rival to the buyer. The idea is that after controlling for factors influential in pricing6_{, ∆}j

l acts as an indicator

of variations in local market power measure, the cost difference between potential competitor and dominant competitor at each location. Using OLS and GMM and via interacting a two level factorial variable, the dummy for first seven months, with market power measures, two different pricing equations are estimated; one for first seven months, and the other for after month seven. To my best knowledge, there is no work with similar methodology in homogeneous product / geographic space setting.

Findings indicate that i) at locations where market power of provider and the closest rival converge, there is a large price difference between suspected collusion period

6_{Controlling for all factors influential in pricing is naturally not possible; however, as part of} future work, I plan to study other confounding factors.

and competition period; ii) at locations where the provider has large market power, price in both periods converge; iii) in suspected competition period, local market power indicator is both linearly and quadratically related to pricing; providers suffer large price cuts to serve buyers that are gradually closer to the closest rival; iv) in suspected collusion period, local market power indicator is positively but only linearly related to price, and the linear relation is much weaker than that in competition. These findings are interpreted as further evidence for a regime switch from collusion to competition. The results also suggest that level of market power each provider has on a buyer is very important in the assessment of the impact of collusion on price, which is explored in detail in Chapter 5.

This chapter is organized as follows: Next section discusses the motivation in proactive detection, and previous episodes of successful detection. Third section introduces the literature, empirical strategy methodology, and contribution of this work to the literature. Fourth section presents the estimations. Final section concludes.